Menu
Menu
Menu
Menu
Background Non-fatal outcomes of disease and injury increasingly detract from the ability of the world’s population to live in full health, a trend largely attributable to an epidemiological transition in many countries from causes affecting children, to non-communicable diseases (NCDs) more common in adults. For the Global Burden of Diseases, Injuries, and Risk Factors Study 2015 (GBD 2015), we estimated the incidence, prevalence, and years lived with disability for diseases and injuries at the global, regional, and national scale over the period of 1990 to 2015. Methods We estimated incidence and prevalence by age, sex, cause, year, and geography with a wide range of updated and standardised analytical procedures. Improvements from GBD 2013 included the addition of new data sources, updates to literature reviews for 85 causes, and the identification and inclusion of additional studies published up to November, 2015, to expand the database used for estimation of non-fatal outcomes to 60 900 unique data sources. Prevalence and incidence by cause and sequelae were determined with DisMod-MR 2.1, an improved version of the DisMod-MR Bayesian meta-regression tool first developed for GBD 2010 and GBD 2013. For some causes, we used alternative modelling strategies where the complexity of the disease was not suited to DisMod-MR 2.1 or where incidence and prevalence needed to be determined from other data. For GBD 2015 we created a summary indicator that combines measures of income per capita, educational attainment, and fertility (the Socio-demographic Index [SDI]) and used it to compare observed patterns of health loss to the expected pattern for countries or locations with similar SDI scores. Findings We generated 9·3 billion estimates from the various combinations of prevalence, incidence, and YLDs for causes, sequelae, and impairments by age, sex, geography, and year. In 2015, two causes had acute incidences in excess of 1 billion: upper respiratory infections (17·2 billion, 95% uncertainty interval [UI] 15·4–19·2 billion) and diarrhoeal diseases (2·39 billion, 2·30–2·50 billion). Eight causes of chronic disease and injury each affected more than 10% of the world’s population in 2015: permanent caries, tension-type headache, iron-deficiency anaemia, age-related and other hearing loss, migraine, genital herpes, refraction and accommodation disorders, and ascariasis. The impairment that affected the greatest number of people in 2015 was anaemia, with 2·36 billion (2·35–2·37 billion) individuals affected. The second and third leading impairments by number of individuals affected were hearing loss and vision loss, respectively. Between 2005 and 2015, there was little change in the leading causes of years lived with disability (YLDs) on a global basis. NCDs accounted for 18 of the leading 20 causes of age-standardised YLDs on a global scale. Where rates were decreasing, the rate of decrease for YLDs was slower than that of years of life lost (YLLs) for nearly every cause included in our analysis. For low SDI geographies, Group 1 causes typically accounted for 20–30% of total disability, largely attributable to nutritional deficiencies, malaria, neglected tropical diseases, HIV/AIDS, and tuberculosis. Lower back and neck pain was the leading global cause of disability in 2015 in most countries. The leading cause was sense organ disorders in 22 countries in Asia and Africa and one in central Latin America; diabetes in four countries in Oceania; HIV/AIDS in three southern sub-Saharan African countries; collective violence and legal intervention in two north African and Middle Eastern countries; iron-deficiency anaemia in Somalia and Venezuela; depression in Uganda; onchoceriasis in Liberia; and other neglected tropical diseases in the Democratic Republic of the Congo. Interpretation Ageing of the world’s population is increasing the number of people living with sequelae of diseases and injuries. Shifts in the epidemiological profile driven by socioeconomic change also contribute to the continued increase in years lived with disability (YLDs) as well as the rate of increase in YLDs. Despite limitations imposed by gaps in data availability and the variable quality of the data available, the standardised and comprehensive approach of the GBD study provides opportunities to examine broad trends, compare those trends between countries or subnational geographies, benchmark against locations at similar stages of development, and gauge the strength or weakness of the estimates available. Funding Bill & Melinda Gates Foundation.
We estimated incidence and prevalence by age, sex, cause, year, and geography using a wide range of updated and standardised analytical procedures. The overall logic of our analytical approach is shown for the entire non-fatal estimation process in figure 1. The appendix provides a single source for detail of inputs, analytical processes, and outputs and methods specific to each cause. This study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) recommendations (methods appendix pp 1, 608–10).17 Analytical flow chart for the estimation of cause-specific YLDs by location, age, sex, and year for GBD 2015 Ovals represent data inputs, square boxes represent analytical steps, cylinders represent databases, and parallelograms represent intermediate and final results. The flow chart is colour-coded by major estimation component: raw data sources, in pink; data adjustments, in yellow; DisMod-MR 2.1 estimation, in purple; alternative modelling strategies, in light green; injury modelling strategy, in dark green; estimation of impairments and underlying causes, in brown; post-DisMod-MR and comorbidity correction, in blue; disability weights, in orange; and cause of death and demographic inputs, in grey. GBD=Global Burden of Disease. TB=tuberculosis. SF-12=Short Form 12 questions. MEPS=Medical Expenditure Panel Surveys. CSMR=cause-specific mortality rate. SMR=standardised mortality ratio. YLDs=years lived with disability. YLLs=years of life lost. IHME=Institute for Health Metrics and Evaluation. The geographies included in GBD 2015 have been arranged into a set of hierarchical categories composed of seven super-regions and a further nested set of 21 regions containing 195 countries and territories. Eight additional subnational assessments were done for Brazil, China, India, Japan, Kenya, Saudi Arabia, South Africa, Sweden, and the USA (methods appendix pp 611–24). For this study we present data at the national and territory level. The GBD cause and sequelae list is organised hierarchically (methods appendix 625–53). At Level 1 there are three cause groups: communicable, maternal, neonatal, and nutritional diseases (Group 1 diseases); non-communicable diseases; and injuries. These Level 1 aggregates are subdivided at Level 2 of the hierarchy into 21 cause groupings. The disaggregation into Levels 3 and 4 contains the finest level of detail for causes captured in GBD 2015. Sequelae of diseases and injuries are organised at Levels 5 and 6 of the hierarchy. The finest detail for all sequelae estimated in GBD is at Level 6 and is aggregated into summary sequelae categories (Level 5) for causes with large numbers of sequelae. Sequelae in GBD are mutually exclusive and collectively exhaustive, and thus our YLD estimates at each level of the hierarchy sum to the total of the level above. Prevalence aggregations are estimated at the level of individuals who might have more than one sequela or disease and therefore are not additive. The cause and sequelae list was expanded based upon feedback after the release of GBD 2013 and input from GBD 2015 collaborators. Nine causes for which non-fatal outcomes are estimated were added: Ebola virus disease, motor-neuron disease, environmental heat and cold exposure, four subtypes of leukaemia, and two subtypes of non-melanoma skin cancer (methods appendix pp 625–53). The incorporation of these changes expanded the cause list from the 301 causes with non-fatal estimates examined in GBD 2013, to 310 causes with non-fatal estimates and from 2337 to 2619 unique sequelae at Level 6 of the hierarchy. At the newly created Level 5 of the hierarchy there were 154 summary sequela categories. The methods appendix (pp 654–61) provides a list of International Classification of Diseases version 9 (ICD-9) and version 10 (ICD-10) codes used in the extraction of hospital and claims data, mapped to GBD 2015 non-fatal causes, impairments, and nature of injury categories. A complete set of age-specific, sex-specific, cause-specific, and geography-specific incidence and prevalence numbers and rates were computed for the years 1990, 1995, 2000, 2005, 2010, and 2015. In this study we focus on trends for main and national results over the past decade, from 2005 to 2015, together with more detailed results for 2015. Online data visualisations at vizhub provide access to results for all GBD metrics. Non-fatal modelling strategies vary substantially between causes. Figure 1 outlines the general process of non-fatal outcome estimation from data inputs to finalisation of YLD burden results; step 3b of that process identifies alternative modelling approaches used for specific causes (methods appendix pp 603, 604). The starting point for non-fatal estimation is the compilation of data sources identified through systematic analysis and extractions based on predetermined inclusion and exclusion criteria (methods appendix p 603). As part of the inclusion criteria, we defined disease-specific or injury-specific reference case definitions and study methods, as well as alternative allowable case definitions and study methods which were adjusted for if we detected a systematic bias. We used 15 types of primary data sources representing disease prevalence, incidence, mortality risk, duration, remission, or severity in the estimation process (oval shapes in figure 1). For this iteration of the study, we updated data searches through systematic data and literature reviews for 85 causes published up to Oct 31, 2015. For other causes, input from GBD collaborators resulted in the identification and inclusion of a small number of additional studies published after January, 2013. Data were systematically screened from household surveys archived in the Global Health Data Exchange, sources suggested to us by in-country experts, and surveys identified in major multinational survey data catalogues and Ministry of Health and Central Statistical Office websites. Case notifications reported to WHO were updated up to and including 2015. Citations for all data sources used for non-fatal estimation in GBD 2015 are provided in searchable form through a new web tool. A description of the search terms used for cause-specific systematic reviews, inclusion and exclusion criteria, and the preferred and alternative case definitions and study methods are detailed by cause in the methods appendix (pp 26–601). Hospital inpatient data were extracted from 284 country-year and 976 subnational-year combinations from 27 countries in North America, Latin America, Europe, and New Zealand. Outpatient encounter data were available from the USA, Norway, Sweden, and Canada for 48 country-years. For GBD 2015, we also accessed aggregate data derived from claims information in a database of US private and public insurance schemes for the years 2000, 2010, and 2012. From the linked claims data, we generated several correction factors to account for bias in health service encounter data from elsewhere, which were largely available to us aggregated by ICD code and by primary diagnosis only. First, for chronic disorders, we estimated the ratio between prevalence from primary diagnoses and prevalence from all diagnoses associated with a claim. Second, we used the claims data to generate the average number of outpatient visits per disorder. Similarly, we generated per person discharge rates from hospital inpatient data in the USA and New Zealand, the only sources with unique patient identifiers available for GBD 2015. In GBD 2013, we calculated a geographical and temporal data representativeness index (DRI) of non-fatal data sources for each cause or impairment. The DRI represents the fraction of countries for which any incidence, prevalence, remission, or mortality risk data were available for a cause. This metric quantifies data availability, not data quality. The overall DRI and period-specific DRI measures for each cause and impairment are presented in the methods appendix (pp 662–68). DRI ranged from 90% for nine causes, including tuberculosis and measles, to less than 5% for acute hepatitis C and the category of other exposures to mechanical forces. Required case reporting resulted in high DRI values for notifiable infectious diseases; the network of population-based registries for cancers resulted in a DRI of above 50%. DRI values ranged from 6·1% in North Korea to 91·3% in the USA. Many high-income countries, as well as Brazil, India, and China, had DRI values above 63%; data availability was low in several countries, including Equatorial Guinea, Djibouti, and South Sudan. In addition to the corrections applied to claims and hospital data, a number of other adjustments were applied including age–sex splitting, bias correction, adjustments for under-reporting of notification data, and computing expected values of excess mortality. In GBD 2013, we estimated expected values of excess mortality from prevalence or incidence and cause-specific mortality rate data for a few causes only, including tuberculosis and chronic obstructive pulmonary disease. In order to achieve greater consistency between our cause of death and non-fatal data, we adopted this strategy systematically for GBD 2015. We matched every prevalence data point (or incidence datapoint for short duration disorders) with the cause-specific mortality rate value corresponding to the age range, sex, year, and location of the datapoint. The ratio of cause-specific mortality rate to prevalence is conceptually equivalent to an excess mortality rate. To estimate non-fatal health outcomes in previous iterations of GBD, most diseases and impairments were modelled in DisMod-MR, a Bayesian meta-regression tool originally developed for GBD 2010 (step 3a in figure 1).18 DisMod-MR was designed to address statistical challenges in estimation of non-fatal health outcomes, and for synthesis of often sparse and heterogeneous epidemiological data. For GBD 2015, the computational engine of DisMod-MR 2.1 remained unchanged, but we substantially rewrote the code that organises the flow of data and settings at each level of the analytical cascade. The sequence of estimation occurs at five levels: global, super-region, region, country, and where applicable, subnational locations (appendix pp 611–24). At each level of the cascade, the DisMod-MR 2.1 computational engine enforces consistency between all disease parameters. For GBD 2015, we generated fits for the years 1990, 1995, 2000, 2005, 2010, and 2015. We log-linearly interpolated estimates for the intervening years in each 5-year period. Greater detail on DisMod-MR 2.1 is available at Global Health Data Exchange and the methods appendix (pp 7–11). In previous iterations of GBD, custom models were created for a short list of causes for which the compartment model underpinning DisMod (susceptible, diseased, and dead) was insufficient to capture the complexity of the disease or for which incidence and prevalence needed to be derived from other data. Step 3b of figure 1 describes the development of custom models with greater detail shown in the methods appendix figure 1B (p 604, and for associated write-ups pp 26–601) for HIV/AIDS, tuberculosis, malaria, cancer, neonatal disorders, infectious diseases for which we derived incidence from seroprevalence data, and infectious diseases for which we derived incidence from cause of death rates and pooled estimates of the case fatality proportion. In GBD 2013, we estimated the country–age–sex–year prevalence of nine impairments (step 4 of figure 1). Impairments in GBD are disorders or specific domains of functional health loss that are spread across many GBD causes as sequelae and for which there are better data to estimate the occurrence of the overall impairment than for each sequela based on the underlying cause. Overall impairment prevalence was estimated with DisMod-MR 2.1 except for anaemia, for which spatiotemporal Gaussian Process regression methods were applied. We constrained cause-specific estimates of impairments, such as in the 19 causes of blindness, to sum to the total prevalence estimated for that impairment. Anaemia, epilepsy, hearing loss, heart failure, and intellectual disability were estimated at different levels of severity. In step 5, sequelae were further defined in terms of severity for 194 causes at Level 4 of the hierarchy (figure 1A). We generally followed the same approach for estimating the distribution of severity as in GBD 2013. For Ebola virus disease, we created a health state for the infectious disease episode with duration derived from average hospital admission times, and a health state for ongoing postinfection malaise and joint problems based on four follow-up studies19, 20, 21, 22 from which we derived an average duration. The health states for the subtypes of leukaemia and non-melanoma skin cancer were the same as the general cancer health states. For motor-neuron disease we accessed the Pooled Resource Open-Access ALS Clinical Trials (PROACT) database containing detailed information on symptoms and impairments for more than 8500 patients who took part in the trials.23 We used the same disability weights as in GBD 2013 (see methods appendix pp 669–94 for a complete listing of the lay descriptions and values for the 235 health states used in GBD 2015). In step 7, we estimated the co-occurrence of different diseases by simulating 40 000 individuals in each geography–age–sex–year combination as exposed to the independent probability of having any of the sequelae included in GBD 2015 based on disease prevalence. We tested the contribution of dependent and independent comorbidity in the US Medical Expenditure Panel Surveys (MEPS) data, and found that independent comorbidity was the dominant factor even though there are well known examples of dependent comorbidity. Age was the main predictor of comorbidity such that age-specific microsimulations accommodated most of the required comorbidity correction. Taking dependent comorbidity into account changed the overall YLDs estimated in the MEPS data by only 2·5% (and ranging from 0·6% to 3·4% depending on age) in comparison to assuming independent comorbidity (methods appendix pp 18–20).24 We report 95% uncertainty intervals (95% UI) for each quantity in this analysis using 1000 samples from the posterior distribution of prevalence and 1000 samples of the disability weight to generate 1000 samples of the YLD distribution. The 95% UI is reported as the 25th and 975th values of the distribution. We report significant changes in disease estimates between countries or over time if the change was noted in more than 950 of the 1000 samples computed for each result. For GBD 2015, we computed age-standardised prevalence YLD rates from the updated world population age standard developed for GBD 2013.25 Less common diseases and their sequelae were included in 35 residual categories (methods appendix pp 695–97). For 22 of these residual categories, estimates were made from epidemiological data for incidence or prevalence. For 13 residual categories, we estimated YLDs by multiplying the residual YLL estimates by the ratio of YLDs to YLLs from the estimates for explicitly modelled Level 3 causes in the same disease category. In GBD 2013, a sociodemographic status variable was computed based on a principal components analysis of income per capita, educational attainment, average age of the population, and the total fertility rate.26 For GBD 2015, we excluded mean age of the population because it is directly affected by death rates. To improve interpretability for GBD 2015, we computed a Socio-demographic Index (SDI) similar to the computation of the human development index.27 In the SDI, each component was weighted equally and rescaled from zero (for the lowest value observed during 1980–2015) to one (for the highest value observed) for income per capita and average years of schooling, and the reverse for the total fertility rate. The final SDI score was computed as the geometric mean of each of the components. SDI ranged from 0·060 in Mozambique in 1987 to 0·978 in District of Columbia, USA, in 2015. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.
